Getting Started
October 19, 2026
Conversion funnel tracking and audience segmentation: see where you lose momentum, and what to fix
Vincent Gaemers
On this page
Key takeaways
- A conversion funnel is an ordered set of steps from first visit to revenue, counted per person or per company, with the drop-off at each step. Most funnels count sessions, stop at 30 days, and end at the browser, so they report drop-offs that never happened and miss the ones that did.
- A funnel on its own is a fact. A funnel split by source is a decision. In our example, Meta brings 40% of visits and 7% of customers; the all-traffic column hides that.
- Every drop-off is a segment: the people who reached a step and didn't take the next one. Name them, and you can retarget them, nudge them, or hand them to sales within minutes.
- The segments that earn their keep are built on the timeline, not the pixel: pricing visits, activation, CRM stage, payments, LTV. A pixel audience can't suppress a customer who closed through sales.
- Funnel, segment, action, measure. Attribution is the fourth step, not the first: it tells you which channels fed the step you fixed.
What is conversion funnel tracking?
Conversion funnel tracking follows people through an ordered set of steps, from the first visit to revenue, and counts how many reach each step and how many leave between steps. The steps are events: Visit, Pricing page, Signup, Activated, Charged for a SaaS product. Visit, Brochure download, Price calculator, Viewing, Signed for a property developer. The drop-off between 2 steps is where you lose momentum, and the people who left are who you talk to next.
That's the idea. Here's why most funnels lie.
3 ways a funnel lies
It counts sessions. A session funnel counts the same person 3 times if they come back 3 times, and calls a lunch break a drop-off. One B2B buyer viewed 51 pages over 6 visits in 90 days before signing. In a session funnel that's 6 visitors, 5 of whom "dropped out".
It stops at 30 days. Most analytics funnels only connect steps inside a window. On a long cycle, the step after "pricing" can be 4 months away. The funnel reports a drop-off at pricing that never happened; the buyer just took their time.
It ends at the browser. The last step a web analytics tool can see is a form or a thank-you page. Deal created, deal won, and the invoice live in the CRM and the billing system. A funnel that ends at "form submitted" tells you who fills in forms, not who pays. In real estate, 1 channel cost €878 per CRM contact and another €128; measured per form fill, they looked alike.
A funnel that tells the truth counts people (and companies, in B2B), spans the whole cycle, and ends at revenue. The rest of this guide is how to build one, how to read it, and what to do with the people who leave.
How it works: 1 timeline, 3 ways to read it
Everything rests on 1 timeline per person, rolled up per company. It holds every visit with its source and campaign, the moment the visitor is identified, the product and CRM events (trial started, deal created, stage changed, won), and the revenue events from Stripe and the CRM.
The funnel, the segments, and attribution are 3 ways of reading that 1 timeline.
| Reading | Question | How it reads the timeline |
|---|---|---|
| Funnel | Where do people leave between steps, and does it differ by source? | An ordered list of events, each person counted once at the furthest step reached |
| Segment | Who is in a given state right now? | A filter on events, properties, CRM stage, and revenue, kept current |
| Attribution | Which touchpoints fed the people who reached the last step? | A credit split across the visits before the conversion |
Because all 3 read the same events, they agree by construction. The 600 signups in the funnel are the 600 whose journeys attribution splits credit for, and the 1,800 people who reached pricing and left are a segment you can retarget 5 minutes later. Build them in separate tools and you get 3 definitions of a conversion and 3 answers.
Building a funnel that tells the truth
Three decisions: which steps, what you count, and how long a step may take.
1. Pick 4 to 6 steps that end at revenue
Fewer steps than you think. Each step is 1 event that marks a real change in intent, and the last one is money. Typical shapes:
| Growth model | Steps | The step that predicts revenue |
|---|---|---|
| SaaS, self-serve | Visit → Pricing → Signup → Activated → Charged | Activated. Signups that never activate are an onboarding problem, not an ad problem. |
| B2B, sales-led | Visit → Content or pricing → Demo booked → Deal created → Won | Demo booked, counted per company. |
| Real estate developer | Visit → Brochure → Price calculator → Viewing → Signed | Price calculator. The strongest intent signal in property; brochure downloads are collectors. |
| Subscription or membership | Visit → Plan page → Checkout started → First payment → Month-2 payment | Month-2 payment. The first payment isn't the value. Use a separate event per product type, so members and one-offs don't share a step. |
| Long-cycle B2C | Visit → Configurator or calculator → Call or appointment → Quote → Purchase | The call. It comes from the CRM or call tracking, not the website. |
Offline steps join from the CRM: calls, viewings, appointments, quotes. If a step only exists in someone's head or a spreadsheet, it isn't a step yet.
2. Count people and companies, not sessions
Each person counts once, at the furthest step they reached, however many visits it took. In B2B, roll colleagues into 1 company: HubSpot showed about 46 contacts at deal-won where there were 9 businesses, and the company funnel was the one that matched the invoices. That takes identify() on every form and group() for the company.
3. Let a step take as long as it takes
Set no window between steps, or set it to your real cycle. A 7-month B2B funnel measured with a 30-day step window shows a wall at step 2. Keep every touchpoint, and let the funnel span months.
In Spectacle, cross-channel funnels are built from any events on the timeline, counted per person or per company, with offline steps from the CRM, and no cap on how long a journey may run.
Reading the drop-off
The funnel on its own is a fact. The funnel split by source is a decision. Here's 1 month for a self-serve SaaS product, illustrative numbers, typical shape.
| Step | All traffic | Google Ads | Meta Ads | Organic |
|---|---|---|---|---|
| Visit | 10,000 | 3,000 | 4,000 | 3,000 |
| Pricing | 2,400 (24%) | 900 (30%) | 600 (15%) | 900 (30%) |
| Signup | 600 (25% of pricing) | 270 (30%) | 120 (20%) | 210 (23%) |
| Activated | 240 (40% of signups) | 135 (50%) | 24 (20%) | 81 (39%) |
| Charged | 60 (25% of activated) | 34 (25%) | 4 (17%) | 22 (27%) |
Read the all-traffic column and the funnel looks healthy: 24% reach pricing, 40% of signups activate. Read it by source and Meta is the problem. It brings 40% of the visits and 7% of the customers, and the damage is at 2 steps: 85% of its visitors never see pricing, and 80% of its signups never activate.
The 3 questions to ask at every step
Who leaves? Split the step by source, campaign, landing page, device, country, or company size. A drop-off that's the same for every source is a page problem. A drop-off that belongs to 1 source is a targeting problem.
How long does the step take? Median days from step to step, by source. If organic takes 9 days from pricing to signup and paid takes 2, the paid visitors are further along when they arrive, or the organic ones are reading. Both are useful.
What do the leavers have in common? Open the people who reached the step and didn't take the next one. The 600 Meta visitors who left at pricing are a list, not a percentage. That list is the start of segmentation.
Meta leaks at 2 steps: pricing and activation
Step-to-step conversion by source.
Illustrative example from this guide.
In the example, the decisions write themselves. The 480 Meta pricing visitors who didn't sign up: retarget on the plan they looked at. The 96 Meta signups who didn't activate: an onboarding sequence, and suppress them from acquisition ads. And the Meta targeting itself: find which campaigns and placements produced the 24 who activated, and cut the rest.
Every drop-off is a segment you can act on
Meta's 4,000 visitors, step by step. The people who left are lists, not percentages.
Illustrative example from this guide.
Audience segmentation
A segment is a saved filter on the timeline: everyone whose events, properties, CRM stage, or revenue match a rule, kept current as people enter and leave. Every drop-off in the funnel is a segment waiting to be named: reached step 3, never reached step 4.
Ad platforms segment on demographics and pixel events. A segment built on the full timeline can use the things that predict revenue, and it can see what happened outside the browser: the deal that closed through sales, the payment that landed in Stripe, the company that Lens identified.
The 7 segments that earn their keep
| Segment | Rule | What you do with it |
|---|---|---|
| Stalled at pricing | Visited pricing 2+ times in 14 days, no signup | Retarget on the plan they viewed; alert sales if the company fits your ICP |
| Signed up, not activated | Signup, no activation event within 7 days | Onboarding sequence and in-product nudges; suppress from acquisition ads |
| Open pipeline | CRM stage between deal created and won | Exclude from cold campaigns, so budget goes to net-new accounts |
| Won | Deal won, or first payment | Suppress everywhere; seed for lookalikes |
| High-LTV customers | 12-month revenue in the top quartile | Lookalike seed on Meta, Google, and OpenAI, filtered by acquisition channel |
| Churned | Subscription ended in the last 90 days | Win-back campaign, separate from acquisition |
| In-market accounts | Company identified by Lens, 3+ visits in 30 days, no CRM record | ABM list on LinkedIn; Slack alert to sales |
The rule can combine anything on the timeline: a page, an event, a campaign, a CRM property, a Stripe amount, a company attribute. That's the whole difference from a pixel audience.
2 rules for segments
Suppress on CRM and billing, not on pixel conversions. A pixel "purchase" fires on a thank-you page. A CRM "won" fires when the deal closes, including the deals sales closed over the phone. Suppress on the second, or you keep paying to reach customers who signed last week.
Sync both ways. A segment that only adds people goes stale in a week. When someone activates, they should leave "signed up, not activated" within minutes, not at the next CSV upload. One-way syncs are for conversion events; two-way syncs are for lists.
In Spectacle, Audience Hub builds segments from website, product, CRM, and revenue signals and keeps them live in Meta Custom Audiences, Google Customer Match, LinkedIn Matched Audiences, and OpenAI, with members added and removed in both directions.
The loop: funnel, segment, action, measure
Put the 2 together and you get a loop that runs every week.
- The funnel finds the step. Split by source, 1 step and 1 source stand out: Meta signups, 80% never activate.
- The segment names the people. Signed up from Meta, no activation in 7 days: 96 people this month, kept current.
- The action goes to the segment. An onboarding sequence to the 96. Suppress them from acquisition. Send "Activated", not "Signup", back to Meta as the conversion, so its algorithm optimizes on the step that predicts revenue.
- Attribution measures the fix. Next month, the funnel shows whether Meta's activation rate moved, and the attribution report shows whether the activated users from Meta turned into revenue. The multi-touch guide covers how that credit is split.
Attribution comes last on purpose. It answers "which channels fed the step we fixed?" If you start with attribution, you get a channel ranking and no idea which step to fix.
What changes when the funnel ends at revenue
4 situations from Spectacle accounts, anonymized, where a funnel or a segment changed a decision.
The funnel that ended at "trial started"
A SaaS founder's funnel stopped at the last event the analytics tool could see. Connecting Stripe added "Charged" and showed whole chunks of payments had been invisible: the cheapest trials, by channel, were worth the least. The funnel moved to a company view on top of the CRM, so the team could see which businesses came in, who paid, and where they dropped off.
Action: end the funnel at revenue, not at the last event your tool happens to see.
Buyers vs brochure collectors
A property developer had hundreds of leads per project and no idea which campaigns brought buyers. Measured per CRM contact instead of per form fill, 1 channel cost €878 per contact and another €128. The price calculator, not the brochure, was the step that predicted a signature; leads carried 21+ touchpoints before it.
Action: make the price calculator the funnel's intent step, cut the expensive channel, and retarget the segment that used the calculator.
The checkout that couldn't tell members from one-offs
A membership business had 80% cart abandonment and no step visibility, and its checkout sent 1 conversion event for memberships and one-off purchases alike, so the ad platforms optimized on the wrong buyers. Separate events per product type gave each its own funnel and its own conversion signal, and 12-month cohort revenue (e.g., all members who joined in March) showed which campaigns brought members who renewed.
Action: 1 event per product type, and judge channels on 12-month revenue, not on first payment.
The signups that never activated
A PLG team saw Google "converting" at twice Meta's rate. Split by step, both channels produced signups at similar cost; Meta's never activated. Retargeting Meta's signups on the activation step, suppressing everyone already activated, and sending the activation event back to Meta as the conversion fixed the number without touching the creative.
Action: optimize the ad platform on the step that predicts revenue, and send that step back as the conversion.
Common funnel and segmentation problems (and how to fix them)
- Too many steps. A 12-step funnel has 11 drop-offs and no decision. Keep 4 to 6 steps, each a real change in intent, the last one revenue.
- Steps that aren't events. "Interested" and "Qualified" are opinions until they're a CRM stage with a date. Define every step as 1 event with 1 definition and 1 destination, and send it with track().
- Session-based counting. The same person counted 3 times, a lunch break called a drop-off. Count per person with identify() on every form, and per company in B2B with group() and company attribution on.
- A step window shorter than the cycle. A wall at step 2 on a 7-month funnel. Set no window, or your real cycle.
- A funnel that ends at the browser. Connect the CRM and Stripe, so the last steps are deal won and paid, and call tracking so phone conversions count. One client found 3x more phone revenue than they knew existed the moment the CRM was connected.
- Segments that only grow. A list that never removes anyone keeps paying for customers. Use two-way list syncs, and suppress on CRM stage and payments, not on pixel events.
- One event for every product. Memberships and one-offs in the same step, so the ad platform optimizes on whoever is cheapest. One event per product type.
- "Everything is direct." Consent walls, device switches, and stripped UTMs file your best traffic under direct, so the funnel by source is wrong before you read it. Route tracking through your own domain with server-side tracking, and keep UTMs consistent (UTM guidelines, or check yours with UTM Scout).
- Tracking that quietly breaks. After a plugin change in 1 account, only about 60% of form submits were still firing, and the funnel looked normal. Write the event plan down, use 1 tag per platform, and check the counts against the CRM monthly.
- Reading the drop-off as the cause. A 60% drop at pricing is where people leave, not why. Open the segment, look at what they have in common, and talk to 10 of them before rebuilding the page.
How to set up funnel tracking and segmentation in 5 steps
Most teams are done in about 30 minutes. Step 2 is the one to spend time on: the event plan is the funnel.
1. Clean 5 CRM fields first
1 currency and 1 amount field per deal, 1 record per company, a defined lead stage, source fields nobody overwrites, and real closed-won dates. The last steps of every funnel live in the CRM, so the CRM decides whether they're true.
2. Write the event plan: 1 event per step
List the 4 to 6 steps as events, each with a definition and a destination. For SaaS the minimum is Customer Created, Trial Started, Subscription Started, Charged; for property it's Brochure, Price Calculator, Viewing, Signed. One event per product type. Send them with track(), revenue in cents. This 1 list feeds the funnel, the segments, and the attribution report.
3. Track every page, identify, and group
Add the JavaScript snippet to every page, or use the Google Tag Manager template, Segment, or the REST API, routed through server-side tracking. Call identify() wherever someone shares contact details, group() for the company, and turn on company attribution. On single-page apps, call page() on every route change.
4. Connect the CRM, payments, and ad accounts
Connect your CRM (HubSpot, Salesforce, Dynamics 365, Odoo), payments (Stripe), call tracking (CallRail), and ad accounts, so deals, payments, calls, and spend land on the timeline on their own.
5. Build 1 funnel and 3 segments, then put them in a meeting
One funnel from visit to revenue, split by source. Three segments: stalled at the step with the biggest drop-off, open pipeline, and won. Sync the first to your ad platforms for retargeting and the other 2 as exclusions, and send the revenue-predicting step back to the platforms as the conversion. Then put the funnel in the weekly growth meeting. Adoption is scheduled, not assumed.
Choosing web analytics software for funnels and segments: 7 checks
| Check | What to ask | GA4 | What to look for |
|---|---|---|---|
| Counting unit | Sessions, people, or companies? | Sessions and users, no companies | People, rolled up into companies for B2B |
| Funnel span | Can a step be months after the last one? | Within its attribution lookback, 90 days at most | Every touchpoint kept; the window set per report |
| Funnel steps | Can a step come from the CRM, billing, or call tracking? | No; on-site events only | Deal stages, payments, and calls as steps |
| Breakdown | Can I split a step by campaign, keyword, landing page, device, company? | Partly, inside the session model | Any dimension on the timeline, no SQL |
| Segment rules | Can a segment use CRM stage and revenue? | On-site events and properties only | Website, product, CRM, and revenue signals |
| Segment sync | Does it push to ad platforms both ways? | Google Ads only, one direction | Add and remove on Meta, Google, LinkedIn, OpenAI |
| Attribution on the same events | Can I see which channels fed the step I fixed, on rules I can explain? | Last click or data-driven | First touch, last touch, linear, U-shaped, side by side |
Price is the last check. Spectacle's funnels, segments, 4 attribution models, company-level funnels, cohort revenue, Lens, and audience syncs are all on the €349 Scale plan. See pricing.
FAQ
What is a conversion funnel in web analytics?
An ordered set of steps from first visit to a conversion, with the count at each step and the drop-off between steps. A funnel that tells the truth counts people rather than sessions, spans the whole buying cycle, and ends at revenue in the CRM or billing system, not at a form.
How many steps should a funnel have?
4 to 6. Each step is 1 event that marks a real change in intent, and the last one is money. More steps means more drop-offs and fewer decisions.
What's the difference between a funnel and a segment?
A funnel counts how many people reached each step. A segment is the list of people in a given state right now, such as everyone who reached pricing twice and never signed up. Every drop-off in a funnel is a segment waiting to be named.
What should audience segments be built on?
On the events that predict revenue: pricing visits, activation, CRM stage, payments, LTV, company fit. Pixel audiences only know what happened in the browser, so they can't suppress a customer who closed through sales or seed a lookalike from the customers who pay the most.
Can GA4 track a conversion funnel across months?
Only within its 90-day attribution lookback, and only on sessions and users, not companies, with no steps from the CRM or billing. A B2B funnel whose last step comes 5 months after the first visit needs a tool that keeps every touchpoint and reads the CRM.
Where does attribution fit?
After the funnel and the segment. The funnel finds the step, the segment names the people, the action goes to the segment, and attribution measures which channels fed the step you fixed. The multi-touch guide covers the models.
How much does funnel tracking software cost?
Spectacle starts at €349 a month with funnels, segments, 4 attribution models, company-level funnels, cohort revenue, Lens, and audience syncs included, with a 14-day free trial and no credit card.
See, per step, where you lose momentum
Stop guessing why people don't convert. Spectacle builds the funnel from visit to revenue, counted per person and per company, splits every step by source, and turns every drop-off into a segment that's live in your ad platforms within minutes. From €349 a month, about 30 minutes to set up.
Start a free 14-day trial or book a demo.
Related reading
- Multi-touch attribution: what it is, 4 models, and how to set it up
- Account-based attribution: how to measure ABM from first touch to pipeline
- Track marketing channels to activated users
- Tracking the features that turn trials into revenue
- Stripe attribution: the cheapest trials were worth the least
- Best Google Analytics alternatives for SaaS
- Cross-channel funnels and Audience Hub
Latest insights
General News
February 27, 2026
Vincent Gaemers